Anssi Kemppainen

Global indoor self-localization based on the ambient magnetic field

There is evidence that animals utilize local anomalities of Earths magnetic field not just for orientation detection but also for true navigation, i.e., some animals are not only able to detect the direction of Earths magnetic field (compass heading), they are able to derive positional information from local cues arising from the local anomalities of Earths magnetic field. Similarly to Earths non-constant magnetic field, the magnetic field inside buildings can be highly non-uniform. The magnetic field fluctuations inside buildings arise from both natural and man-made sources, such as steel and reinforced concrete structures, electric power systems, electric and electronic appliances, and industrial devices. Assuming that the anomalities of the magnetic field inside a building are nearly static and they have sufficient local variability, the anomalies provide a unique magnetic fingerprint that can be utilized in global self-localization. Based on the evidence presented in this article it can be argued that this hypothesis is valid. In this article, a Monte Carlo Localization (MCL) technique based on the above hypothesis is proposed. The feasibility of the technique is demonstrated by presenting a series of global self-localization experiments conducted in four arbitrarily selected buildings, including a hospital. The experiment setup consists of a mobile robot instrumented with a 3-axis magnetometer and a computer. In addition to global robot self-localization experiments, successful person self-localization experiments were also conducted by using a wireless, wearable magnetometer. The reported experiments suggest that the ambient magnetic field may remain sufficiently stable for longer periods of time giving support for self-localization techniques utilizing the local deviations of the magnetic field.

A global self-localization technique utilizing local anomalies of the ambient magnetic field

Magnetic field fluctuations in modern buildings arise from both natural and man-made sources, such as steel and reinforced concrete structures, electric power systems, electric and electronic appliances, and industrial devices. If the anomalies of the magnetic field inside the building are nearly static and they have sufficient local variability, they provide a unique magnetic fingerprint that can be utilized in global self-localization. In this article, a Monte Carlo Localization (MCL) technique based on this hypothesis is proposed. The feasibility of the technique is demonstrated by presenting a series of global localization experiments conducted in four arbitrarily selected buildings, including a hospital. The experiment setup consists of a mobile robot instrumented with a 3-axis magnetometer and a computer. In addition, successful human self-localization experiments were conducted by using a wireless wearable magnetometer. The reported experiments suggest that the ambient magnetic field may remain sufficiently stable for longer periods of time, giving support for self-localization techniques utilizing the local deviations of the field.

Simultaneous localization and mapping using ambient magnetic field

In this paper we propose a simultaneous localization and mapping (SLAM) method that utilizes local anomalies of the ambient magnetic field present in many indoor environments. We use a Rao-Blackwellized particle filter to estimate the pose distribution of the robot and Gaussian Process regression to model the magnetic field map. The feasibility of the proposed approach is validated by real world experiments, which demonstrate that the approach produces geometrically consistent maps using only odometric data and measurements obtained from the ambient magnetic field. The proposed approach provides a simple, low-cost, and space-efficient solution for solving the SLAM problem present in many domestic and swarm robotics application domains.

Magnetic field-based SLAM method for solving the localization problem in mobile robot floor-cleaning task

In this paper we present a SLAM method based on indoor magnetic field anomalies and measure the acquired map quality in the context of the localization problem present in mobile robot floor-cleaning scenarios. According to our real-world robot experiments in different environments, it appears that most modern buildings have sufficient magnetic field variation to make the method applicable in mobile robot floor-cleaning tasks. We show that our method can be used to acquire maps that are accurate enough to be utilized in the robot coverage problem, thus reducing over-cleaning. We use Gaussian Processes to model the magnetic field and a Rao-Blackwellized Particle Filter to estimate the pose distribution of the robot. Because magnetic field anomalies are not correlated to typical features used in localization, our method can handle many situations in which other methods fail. The minimalistic sensory requirements of our method make it a very viable alternative for low-cost domestic robots.

A geomagnetic field based positioning technique for underground mines

A geomagnetic field based positioning technique is proposed for underground mining environments. The proposed technique utilizes the anomalies of the geomagnetic field present in underground environments. The main source of the magnetic anomalies is the complex distribution of metallic minerals such as iron ore. The distribution of metallic minerals produces unique spatial magnetic patterns in underground mines which can be utilized for positioning. Preliminary results are presented using the data collected from Pyhäsalmi, which is an underground copper and zinc mine located in central Finland. The data used in the experiments were collected from tunnels located approximately 1400 meters below the surface. The obtained results suggest that the proposed positioning technique can provide pose estimates with an accuracy of ≈1.5meters. The proposed technique can potentially provide a robust, and a cost-efficient positioning solution for underground environments with minor infrastructure requirements

Near-optimal SLAM exploration in Gaussian processes

In this paper we examine near-optimal SLAM exploration in Gaussian processes. We propose a submodular sensing quality function that extends studies from discrete sensor placement to an autonomous sampling scheme where sensing sites must be visited frequently. This is beneficial in the SLAM context, where sensing sites themselves bear uncertainties. Also in time-critical applications, we have to balance modeling accuracy against sensing time, which introduces noisy samples with only limited replications at each site.

An Infrared Location System for Relative Pose Estimation of Robots

In this work we present an infrared location system for relative pose (position and orientation) estimation in a multi-robot system. Pose estimates are essential for tasks like cooperative simultaneous localization and mapping (C-SLAM), and formation control. In simultaneous localization and mapping (SLAM) relative pose estimates enable more accurate and less time-consuming map building. Respectively, formation control requires accurate pose estimates of other robots to enable robot cooperation in required formation. To address these challenging tasks for small-sized robots, we present a small-sized infrared location system with low current consumption. In the location system, robots use intensity and bearing measurements of received infrared signals to estimate the positions of other robots in polar coordinates. In addition, each robot has a unique modulation frequency from which they are recognized. The location system performs position estimation by rotating a beam collector at constant rotation speed and by measuring the bearing and intensity of the received signal. Infrared signals are received through a small aperture in the beam collector enabling accurate bearing measurements. In order to maximize the measurement range, infrared radiation is reflected sideways into a uniform zone using a conical mirror. Experiments were performed in a group of three robots with a measurement range of up to 3 m while the maximum number of robots was eight. The location system implemented enables relative position estimation among a group of small-sized robots without exchanging position estimates. This is advantageous since the robots are able to maintain formation also in the absence of a radio link.

Magnetic field SLAM exploration: Frequency domain Gaussian processes and informative route planning

In this paper, we consider magnetic field SLAM exploration using a mobile robot with a magnetometer and wheel encoders. We propose computationally feasible solutions to model magnetic fields using frequency domain Gaussian processes. In addition, we propose a path planning algorithm to efficiently collect a given level of accuracy for magnetic field models. The path planning is based on partition of shortest paths into blocks with similar information content and implementing depth-first search among these blocks. Finally, we propose an exploration-exploitation algorithm enabling real-world mobile robot SLAM exploration solutions with motion uncertainties. SLAM is presented with Rao-Blackwellized particle filters where robots path hypothesis are presented with particles together with a separate magnetic field model for each particle. We conducted SLAM exploration experiments using real magnetic field data in a simulated environment. Simulation parameters were tuned to approximate ICreate robots motion uncertainties and MicroMag3 magnetometers sensor noise, together with the robots inclination uncertainties. Simulations demonstrated for the first time that we were able to build actual magnetic field SLAM exploration. The results indicate that, with frequency domain Gaussian processes, we are able to obtain desirable convergence of path distribution, although, with the selected particle filter SLAM approach, the localization accuracy was not desirable.

Monty Hall Particle filter: A new method to tackle predictive model uncertainties

This paper proposes a simple adaptive weight computing method for particle filters that utilizes knowledge about predictive model uncertainty. In each time step the particles are assigned into subsets based on the corresponding uncertainty estimates. The weights are then updated based on accumulated subset-inclusion and likelihood information using a discrete set of measurement likelihood functions. By controlling the aggressiveness of the weight computing, the method strives to achieve faster convergence without losing robustness to model errors. Two localization experiments are conducted to verify that the method has a clear advantage over particle filters with single likelihood function. In the first experiment we use synthetic Gaussian Process data. In the second experiment real indoor magnetic field data with very coarse interpolation and uncertainty approximation is used to verify the methods effectiveness in real-world scenarios. One of the main advantages of the proposed method is that despite its flexibility, it adds only little implementational or computational overhead to conventional particle filters.

Smart system for distributed sensing

This paper describes ongoing work for implementing a distributed multi-robot system for measuring various physical properties of the environment in a coordinated manner. Modeling spatial distributions of physical quantities such as temperature, illumination, humidity, gas concentrations, or magnetic flux provides an opportunity to observe how these distributions change in time, how to utilise the information captured by the distributions, and to determine how to adjust the distributions by modifying the environment. The information provided by the models could be useful in numerous applications such as in dynamic optimisation of the heating system of a building for saving energy, in monitoring the air quality in various parts of hospitals and in monitoring the water quality of lakes and rivers (e.g. pH and oxygen levels). Some of the modeled distributions or maps may also carry information which can be used in mobile computing. This article describes the physical robots, which act as mobile measurement instruments in addition to approaches used to model spatial distributions by utilising Gaussian processes, and to coordinate the multi-robot system during the modeling. Preliminary experiments are also presented demonstrating optimal spatial sampling, optimization of measurement paths of multiple robots, mobile robot self-localisation based on the ambient magnetic field of the environment and multi-robot coordination.